Articleshttps://doi.org/10.1038/s41561-018-0133-5

Tall Amazonian forests are less sensitive to precipitation variabilityFrancesco Giardina1,2, Alexandra G. Konings   3, Daniel Kennedy4, Seyed Hamed Alemohammad   4, Rafael S. Oliveira   5,6, Maria Uriarte7 and Pierre Gentine4,8*

1Department of Earth and Environmental Engineering, Columbia University, New York, NY, USA. 2École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland. 3Department of Earth System Science, Stanford University, Stanford, CA, USA. 4Department of Earth and Environmental Engineering, Columbia University, New York, NY, USA. 5Departamento de Biologia Vegetal, Universidade Estadual de Campinas, Campinas, Brazil. 6School of Plant Biology, University of Western Australia, Perth, Western Australia, Australia. 7Department of Ecology, Evolution and Environmental Biology, Columbia University, New York, NY, USA. 8The Earth Institute, Columbia University, New York, NY, USA. *e-mail: pg2328@columbia.edu

© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

SUPPLEMENTARY INFORMATION

In the format provided by the authors and unedited.

NATURe GeOScieNce | www.nature.com/naturegeoscience

1

Supplementary Material Modelling study experiment description

Our study includes two simulation experiments with a plant hydraulic model to support the observational evidence that taller trees are less sensitivity to variability in precipitation. Since it is difficult to estimate variations in plant hydraulic traits across the Amazon, we do not try to explicitly model any particular scenario or location. Instead, we employ two simple experiments designed to isolate – and illustrate – the role of tree height on the sensitivity of photosynthesis to water availability. Experiment 1 is an extended drydown, repeating the same single day forcing, sixty times. Forcing data was based on the average diurnal cycle of September, 2001 at the Caxiuanã flux tower site (Fluxnet BR-Cax site). An abbreviated model description is provided in the main text (Methods) and the full model code and forcing data are available online at github.com/djk2120/talltrees. We run the soil-plant-atmosphere continuum model for two cases, representing a taller and shorter tree. The only parameters that differ between the cases are: tree height (z), rooting depth (Zr), and the water potential at 50% loss of stomatal conductance (p50). Table S1: Modeling experiment hydraulics parameters

Units Short Tall z Tree height meters 15 40 Kmax Maximum hydraulic conductivity.

Leaf area basis. Unit tree height (m). mmol⋅MPa-1⋅s-1⋅m-1 220 220

Zr Rooting depth meters 2 3 p1 Xylem vulnerability onset MPa -1 -1 p2 Point of 100% loss of xylem

conductivity MPa -4 -4

p50 Point of 50% loss of stomatal conductance

MPa -1.5 -2.5 a Shape fitting parameter for loss of

stomatal conductance (-) 61 61

1Value from Xu et al. 2016 for seasonally dry tropical forest Experiment 2 is the same as Experiment 1, but we remove the bucket model for the tall tree

case. Instead, we force the simulation with the soil moisture from the shorter case, in Experiment 1. This allows us to distinguish the effect of rooting depth from physiological controls.

For Experiment 3, we apply an atmospheric drydown to test the sensitivity of short vs. tall trees to changes in vapour pressure deficit. Over a 60-day drydown, we decrease relative humidity each day by a fixed value (ΔRH = -0.9% per day, which corresponds to -38 Pa per day vapour pressure deficit at midday) relative to the daily forcing from Experiment 1. We use the same hydraulics parameters as Experiment 1, but a constant soil potential forcing of -0.2 MPa, in order to empahisze the role of vapour pressure deficit. Modelling study results

The taller tree case is less sensitive to the drydown. This is true in both experiments as seen in Supplementary Figure 4b,c. Cumulative photosynthesis on day 60 is 76.4% of day 1 photosynthesis for the shorter tree case (same for both experiments). In Experiment 1, the taller trees maintained 94.1% of the day 1 photosynthesis. This is largely due to the effect of deeper rooting, as the shorter trees experienced more negative soil potentials, as low as -1.34 MPa versus only -0.79 for the taller case (Fig S5a).

Experiment 2 eliminates the effect of deep roots, by using the soil potential forcing of the short tree for the tall case. Likewise, the parameter values chosen lead to comparable anisohydricity (defined as the slope of midday leaf water potential to soil water potential), with σ=0.74 for the short trees and σ=0.72 for the tall trees (Supplementary Figure 5b). In Experiment 2, without deeper rooting, tall trees are still less sensitive to the drydown than the shorter trees, maintaining 83.1% of

2

the day 1 photosynthesis (Supplementary Figure 4b). This experiment is used to highlight a second mechanism for tall forest resilience to precipitation variability unrelated to rooting depth and the anisohydricity metric, where taller trees are less sensitive to drops in soil potential, similary to our observations (see main text).

Indeed, this can be explained by the change in midday minus predawn leaf water potential (Δ𝜓) induced by the drydown. In the shorter tree case, the potential drop Δ𝜓 on day 1 is 0.63 MPa (Supplementary Figure 6a). On day 60, the drop is 0.34 MPa. This represents a loss of 45% (0.28MPa) of the day 1 potential drop. For the taller tree case, using the same soil potential forcing (Experiment 2), the day 1 potential drop is -1.55 MPa. This derives from the lower conductance for the taller tree case, due to a longer xylem path and viscous dissipation (Darcy’s law). The tall tree day 60 potential drop Δ𝜓 is -1.24 MPa (Supplementary Figure 6b), representing a loss of 19.7% (0.31 MPa) of the day 1 potential drop. Though the absolute change in potential drop is comparable for the tall and short trees (due to the nearly equivalent anisohydricity), this relative change is larger for the shorter tree. Transpiration supply is proportional to the gradient in water potential, whereby the effect of drydown is larger for the shorter trees, due to the relatively large loss in potential gradient.

The taller tree case is more sensitive to atmospheric demand, Supplementary Figures 4a and 7. Indeed, the taller tree experiences more stress on the first day of the drydown around midday, which is associated with high transpiration demand due to high VPD and photosynthesis, while soil water potential has not had the time to change. To further explore this, we carried out another experiment simulating the atmospheric drydown only, increasing VPD over the course of 60 days (Experiment 3). The midday gross photosynthesis was more sensitive to higher VPD in the taller tree case, decreasing by 2.58 µmol/m2/d as compared to a drop by just 1.45 µmol/m2/d in the shorter tree case (Supplementary Fig. 7). The overall sensitivity for both to VPD is relatively small, because we prescribed well-watered conditions (constant soil potential of -0.2 MPa), to decompose the soil and atmospheric drying effects.

3

Supplementary Figures

Supplementary Figure 1 | Variations of SIF sensitivity to VPD and precipitation binned by tree height (without 2015) and anisohydricity. a,b Partial correlation coefficients for the sensitivity of mean SIF to mean precipitation (blue line) and maximum VPD (red line), binned by tree height, without the 2015 El-Niño year (a) and binned by anisohydricity (b). Normalized yearly values were used for every variable. c, Spatial pattern of anisohydricity in the Amazon. Only tropical rainforests are shown d, Correlation coefficients between precipitation with max VPD (light blue line) and mean VPD (dark red line), binned by tree height.

4

Supplementary Figure 2 | Correlation of mean interannual precipitation with canopy characteristics and precipitation. a, Scatter plot between precipitation and tree age. b, Scatter plot between precipitation and aboveground biomass. c, Scatter plot between precipitation and anisohydricity. d, Scatter plot between mean interannual precipitation and the interannual variability of mean precipitation. Coefficients of determinations are shown in every plot. The significance of correlation was calculated using a t-test with a significance level of 0.05. All correlations are significant at p < 0.001.

5

Supplementary Figure 3 | Rooting depth dependence on soil type. Boxplot of the retrieved rooting depth51, as a function of soil type. The soil types are: 1 Argisol 18 Latosol 2 Argisol 19 Neosol 3 Cambisol 20 Nitosol 4 Cambisol 21 Nitosol 5 Chernosol 22 Vertisol 6 Chernosol 23 Planosol 7 Chernosol 24 Planosol 8 Spodosol 25 Plinthosol 9 Glevsol 26 Plinthosol 10 Glevsol 27 Plinthosol 11 Glevsol 28 Vertisol 12 Latosol 32 Argisol 13 Latosol 33 Argisol 14 Latosol 34 Argisol 15 Luvisol 48 Vertisol 16 Neosol 50 Neosol 17 Neosol

6

Supplementary Figure 4 | Model response of tall vs. short forest gross primary productivity to drydown. Response of the soil-plant-atmosphere-continuum (SPAC) model to 60-day simulated drydown for: (a) Day 1 of Experiment 1, which featured deeper rooting for the tall forest, (b) Day 60 of Experiment 1, (c) Day 60 of Experiment 2, which used identical soil potential forcing for both forests. Blue lines are for the short forest simulation (height set to 15 meters), and the red lines are for the tall forest simulation (height set to 40 meters). Tall forests are more sensitive to atmospheric drying than shorter forests, which are more sensitive to soil drying (both with and without the effect of rooting depth).

0 6 12 18 24Hour

0

5

10

15

20

25G

PP (

mol

m-2

s-1

)Exp1, Day1

(a)

shorttall

0 6 12 18 24Hour

0

5

10

15

20

25Exp1, Day60

(b)

0 6 12 18 24Hour

0

5

10

15

20

25Exp2, Day60

(c)

7

Supplementary Figure 5 | Model leaf water potential response of tall vs. short forest to simulated soil drydown. Midday leaf water potential (relative to day 1) versus soil water potential for: (a) Experiment 1, which featured deeper rooting for the tall forest, and (b) Experiment 2, which used identical soil potential forcing for both forests. Blue points are from the short forest simulation (height set to 15 meters), and the red points are from the tall forest simulation (height set to 40 meters). Anisohydricity for the two forests in these simulations are comparable, with σ=0.74 for the short trees and σ=0.72 for the tall trees (Experiment 2).

-1.5 -1 -0.5 0Soil potential (MPa)

-1

-0.8

-0.6

-0.4

-0.2

0M

idda

y le

af -

leaf

0 (MPa

)Experiment 1

(a)

shorttall

-1.5 -1 -0.5 0Soil potential (MPa)

-1

-0.8

-0.6

-0.4

-0.2

0Experiment 2

(b)

8

Supplementary Figure 6 | Model response of intra-day potential drop to a simulated soil drydown. Experiment 2 (identical soil potential forcing for short and tall forests) time-series of leaf water potential for (a) the short forest simulation, and (b) the tall forest simulation. Day 1 is plotted with the solid lines, and Day 60 is plotted with the broken lines. The short forest loses a larger percentage of the Day 1 intra-day potential drop due to drydown as compared to the tall forest.

0 6 12 18 24Hour

-3

-2.5

-2

-1.5

-1

-0.5

0Le

af p

oten

tial (

MPa

)Short z=15m

(a)Day1Day60

0 6 12 18 24Hour

-3

-2.5

-2

-1.5

-1

-0.5

0Tall z=40m

(b)

9

Supplementary Figure 7 | Model response of midday gross photosynthesis to a simulated atmospheric drydown. Experiment 3, midday GPP response to increasing VPD over a 60-day atmospheric drydown with constant soil potential forcing. Blue points are from the short forest simulation (height set to 15 meters), and the red points are from the tall forest simulation (height set to 40 meters). Tall trees are more sensitive to increasing VPD.

2 2.5 3 3.5 4VPD (kPa)

20

21

22

23

24

GPP

( m

ol m

-2 s

-1)

Exp3: midday GPP vs. VPD

ShortTall

10

Supplementary Figure 8 | Comparison between different water stress indicators. Correlation between mean annual precipitation and cumulative water deficit (CWD) – top, and correlation between mean annual precipitation and standardized precipitation minus evaporation index (SPEI) – bottom.

11

Supplementary Figure 9 | Variations of SIF sensitivity to VPD and precipitation binned by tree height and precipitation, including shortwave radiation. a,b Partial correlation (a) and linear regression (b) coefficients for the sensitivity of mean SIF to mean precipitation (blue line) and maximum VPD (red line), binned by tree height. c,d Partial correlation (c) and linear regression (d) coefficients for the sensitivity of mean SIF to mean precipitation (blue line) and maximum VPD (red line), binned by mean interannual precipitation. Normalized yearly values were used for every variable. Confidence intervals are calculated by bootstrapping across each pixel with n = 2,000. The standard deviation of the n coefficients is used as confidence interval.

12

Supplementary Figure 10 | Interannual variability (in percent) of SIF, precipitation, VPD and shortwave radiation between 2007 and 2015. a, Interannual variability of mean SIF. b, Interannual variability of mean precipitation. c, Interannual variability of maximum VPD. d, Interannual variability of mean shortwave radiation. The interannual variabilities are calculated by dividing the interannual standard deviation by the interannual mean. Only tropical rainforests are shown.

13

Supplementary Figure 11 | Scatter plot of maximum VPD, mean shortwave radiation and mean precipitation with mean SIF for three sample bins of tree height, coloured by pixel. a,d,g, Scatter plot between VPD and SIF. b,e,h, Scatter plot between precipitation and SIF. c,f,i, Scatter plot between shortwave radiation and SIF. Normalized yearly values were used for every variable. Coefficients of determinations are shown in every plot.

14

Supplementary Figure 12 | Variations of SIF sensitivity to shortwave radiation binned by tree height and precipitation. a,b Partial correlation (a) and linear regression (b) coefficients for the sensitivity of mean SIF to mean shortwave radiation, binned by tree height. c,d Partial correlation (c) and linear regression (d) coefficients for the sensitivity of mean SIF to mean shortwave radiation, binned by mean interannual precipitation. Normalized yearly values were used for every variable. Confidence intervals are calculated by bootstrapping across each pixel with n = 2,000. The standard deviation of the n coefficients is used as confidence interval.

15

Supplementary Figure 13 | Scatter plot of climatic drivers for three sample bins of tree height, coloured by pixel. a,d,g, Scatter plot between maximum VPD and mean shortwave radiation. b,e,h, Scatter plot between mean precipitation with yearly maximum VPD. c,f,i, Scatter plot between mean precipitation with mean shortwave radiation. Normalized yearly values were used for every variable. Coefficients of determinations are shown in every plot.

16

Supplementary Figure 14 | Rooting depth estimate based on a fusion between model and groundwater observations51.